Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/ashishgituser/bunkervm/copilot-instructionsgit clone --depth 1 https://github.com/ashishgituser/bunkervmWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.02299 | $0.02299 |
| Opus 5 | $0.01149 | $0.01149 |
| Sonnet 5 | $0.00460 | $0.00460 |
| Haiku 4.5 | $0.00230 | $0.00230 |
Grade B, and why
bunkervm copilot-instructions.md scanned grade B with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
- **vsock** (default): Host ↔ VM via Firecracker UDS. Zero network config, no sudo needed. Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
- **`tools.py`** — Single `execute(cmd)` function using `subprocess.run()`. The entire OS is the tool — no hardcoded tool list. How it starts
The opening of the file, as written. The whole thing — 142 lines — stays where its author put it; the contents beside it link to each section on GitHub.
BunkerVM — Copilot Instructions
Project Overview
BunkerVM is a Python library (pip install bunkervm) that boots disposable Firecracker microVMs for AI agents to execute untrusted code with hardware-level (KVM) isolation. The core loop: boot VM (~3s) → execute code → destroy VM. Only dependency is mcp.
Architecture
Host side (bunkervm/) Guest side (rootfs/bunkervm/)
───────────────────── ────────────────────────────
cli.py / __main__.py → vm_manager.py → [Firecracker process]
↕ vsock ↕
sandbox_client.py ←→ exec_agent.py (HTTP over vsock)
↑
mcp_server.py / integrations/
vm_manager.py— Firecracker process lifecycle (start/stop/restart), TAP networking setup, rootfs copies. Communicates via vsock UDS (/tmp/bunkervm-vsock.sock).sandbox_client.py— Stdlib-only HTTP client that speaks the Firecracker vsock handshake protocol (CONNECT <port>\n→OK <port>\n→ HTTP). No external HTTP libs.rootfs/bunkervm/exec_agent.py— Runs inside the VM. Zero-dependency HTTP server (stdlib only, must work on Alpine/musl). Endpoints:/exec,/read-file,/write-file,/list-dir,/health,/status.mcp_server.py— Exposes sandbox operations as MCP tools viaFastMCP. Global state set by__main__.pybefore server starts.integrations/base.py—BunkerVMToolsBaseprovides 6 shared tool implementations (_run_command,_write_file,_read_file,_list_directory,_upload_file,_download_file). Framework adapters (langchain.py,openai_agents.py,crewai.py) wrap these with framework-specific decorators — never duplicate tool logic in adapters.safety.py— Regex-based command classifier (READ/WRITE/SYSTEM/DESTRUCTIVE/BLOCKED). Defense-in-depth only; the VM is the real isolation boundary.bootstrap.py— Auto-downloads Firecracker bundle to~/.bunkervm/bundle/on first run. Falls back tobuild/dir in dev mode.config.py— Layered config: CLI args > env vars (BUNKERVM_*) >bunkervm.toml> built-in defaults.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 142 lines · 2,299 tokens per session scan B e1ad73ae6b1c
bunkervm copilot-instructions.md is an instructions file published in the GitHub repository ashishgituser/bunkervm (0 stars, last pushed 14d ago), licensed MIT. It adds 2,299 tokens to every session, about $0.0115 per session on Opus 5. A static security scan graded it B with 2 findings (asks for root, runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
mitos CLAUDE.md
Instructions for mitos-run/mitos, covering claude.md, project overview, operating principles, commands and architecture.
oring CLAUDE.md
Instructions for hexproofdev/oring, covering claude.md, the discipline, hard constraints and standing rejections.
arcbox CLAUDE.md
Instructions for arcboxlabs/arcbox, covering repository guidelines for coding agents, project overview, performance targets, platform priority and project structure.
codex-autoresearch AGENTS.md
Instructions for TheGreenCedar/codex-autoresearch, covering agents.md, purpose and scope, canonical sources, stable product boundaries and source and package boundaries.
autoharness AGENTS.md
Instructions for tigerless-labs/autoharness: This repo uses Open Knowledge — collaborative markdown via MCP. .open-knowledge/config.yml (with optional /.open-knowledge/config.yml; CLI/env may override) is the path contract: content.dir is the root for relative paths; content.include lists globs that add markdown…
mosoo AGENTS.md
Instructions for langgenius/mosoo, a project described as: The open-source Agent Gallery and Gateway for Codex, Claude Agent SDK, and OpenCode. Developers publish an Agent once behind one HTTP API; users run it in an isolated cloud sandbox—no local harness setup, compute, or runtime infrastructure.